You are on page 1of 6

Preliminary system and hardware design for Quori,

a low-cost, modular, socially interactive robot


Andrew Specian Nick Eckenstein Mark Yim
The University of Pennsylvania The University of Pennsylvania The University of Pennsylvania
Philadelphia, PA Philadelphia, PA Philadelphia, PA
aspecian@seas.upenn.edu neck@seas.upenn.edu yim@seas.upenn.edu

Ross Mead Braden McDorman Simon Kim


Semio Semio The University of Pennsylvania
Los Angeles, CA Los Angeles, CA Philadelphia, PA
ross@semio.ai braden@semio.ai simon.kim@i-k-studio.com

Maja Matarić
University of Southern California
Los Angeles, CA
mataric@usc.edu
ABSTRACT necessary hardware and software to provide a ready, “out of the
This paper presents progress on Quori, a low-cost socially interac- box” experience. Our hope is that Quori will provide a hardware
tive robot platform comprised of an upper-body humanoid with platform that will enable computing researchers to enter the field of
a rear projection head and two gesturing arms on a mobile base. HRI and develop and test algorithms, as well as conduct statistically
The robot’s hardware is introduced and features are explained. The significant user studies by deploying systems in the real world and
modularity, expandability, customizability, and affordability of the collecting user data to inform further computing research in HRI.
design are discussed. The information provided in this paper is By working with the HRI community to identify the most im-
meant to generate feedback for the final design of the robot, which portant hardware capabilities for a socially interactive robot, some
will be used to produce and award copies of the robot to select optimization can be achieved to lower cost and maximize function-
research groups for both in-lab and “in the wild” deployments. ality. The input of the HRI community on Quori’s design process
was collected via on-line surveys, hosted workshops, and confer-
KEYWORDS ence presentations. The data collected from our quorum (a diverse
group of researchers in the broader HRI community) directed our
socially interactive robots, non-contact HRI, low-cost hardware
design decisions for Quori’s hardware and software. This paper
ACM Reference Format: introduces the hardware and possible functions and uses of Quori.
Andrew Specian, Nick Eckenstein, Mark Yim, Ross Mead, Braden McDor- The analysis and discussion of survey data and the design process,
man, Simon Kim, and Maja Matarić. 2018. Preliminary system and hardware
as well as the integration of the hardware and low-level software,
design for Quori, a low-cost, modular, socially interactive robot. In HRI ’18
are beyond the scope of this paper.
Workshop: 2018 ACM/IEEE International Conference on Human-Robot Inter-
action Workshop on “Social Robots in the Wild”, March 5–8, 2018, Chicago, IL, The goal of this paper is to capitalize on a unique interaction
USA. ACM, New York, NY, USA, 6 pages. https://doi.org/10.1145/nnnnnnn. with the HRI community to receive our final round of feedback on
nnnnnnn the hardware direction and robot capabilities. This work is part
of a National Science Foundation Computing Research Infrastruc-
1 INTRODUCTION ture grant that will be awarding Quori robots to eight university
researchers with releavnt HRI research plans.
Hardware platforms for socially interactive robotics can be limiting
because of its cost or lack of functionality. This paper presents the
hardware for “Quori”, a novel, affordable, socially interactive robot
platform for enabling non-contact human-robot interaction (HRI) 2 SYSTEM OVERVIEW
research in both in-lab and “in the wild” experimental settings. In
its final form, the platform will be a complete package with all Quori consists of an expressive upper body attached to a omnidirec-
tional mobile base (Figure 1). Each part of the system is described
Permission to make digital or hard copies of part or all of this work for personal or in detail in Section 3. We included design considerations for sen-
classroom use is granted without fee provided that copies are not made or distributed sors needed for HRI (Figure 1, right), including sensing the robot’s
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for third-party components of this work must be honored. internal state via motor encoders, as well as a laser range finder,
For all other uses, contact the owner/author(s). camera, and microphones to sense the environment and user(s). An
HRI ’18 Workshop on “Social Robots in the Wild”, March 5, 2018, Chicago, IL, USA
example networking diagram for controlling the robot is shown in
© 2018 Copyright held by the owner/author(s).
ACM ISBN 978-x-xxxx-xxxx-x/YY/MM. Figure 2.
https://doi.org/10.1145/nnnnnnn.nnnnnnn
HRI ’18 Workshop on “Social Robots in the Wild”, March 5, 2018, Chicago, IL, USA A. Specian et al.

an adult human head. The RAF yields an expressive, customizable,


and natural for human-robot interaction [2].

Figure 3: Quori’s head hardware CAD (left) and finished


head module (middle). Sample warped image to be sent to
the projector (top-right) and the final product displayed on
the sphere (bottom-right).
Figure 1: Overview of Quori’s components (left) and sensing
capabilities (right).
Software to map the spherical surface to the projected image and
an intuitive interface for researchers to program and control the
face is needed to display faces properly on the sphere or another
surface. Our mapping algorithm maps pixels on the sphere to pixels
in an image to be sent to the projector (Figure 3, bottom). This is
used to create expressive faces.
The projector is an AXAA P5 currently (2017) priced at $290USD.
It was selected because of its relatively low cost, brightness, small
achievable image size, about 3x5 inches at a short distance of 8
inches, and other properties2 . The resolution of the sphere’s surface
Figure 2: Simple network diagram for Quori.
is not uniform; it is dense near the top and sparse toward the
neck. To provide an idea of the resolution available, the least dense
3 HARDWARE OVERVIEW equatorial line is about 200 pixels. The projector is able to create a
color image that is visible in most illuminated indoor environments
Our hardware design approach consists of three key aspects: 1) where there is no sunlight saturation (Figure 8). This module is
verified utility through iterations with the HRI community to iden- easily detachable by its four mounting holes and can be replaced,
tify desired features; 2) capability using recent technology break- as discussed in Section 3.4.
throughs for low-cost and targeted feature inclusion; and 3) longevity
of impact through development of modular interface standards. This 3.2 Gesturing Arms
section covers the current state of the robot’s design, leading into
the final iteration, which will take place in Summer 2018. At that Gestures are a key part of natural communication in social inter-
time, we will begin to produce multiple copies of the robot for action. Our arm design is low-cost, modular, safe, and expandable.
distribution to the HRI research community. Our 2-DOF shoulder module (Figure 4) is based on a design by
Whitney and Hodgins [12]; however, our design differs in that we
3.1 Spherical Projection Head use 3D-printed bevel gears instead of a capstan cable drive, and
we chose to not gravity-compensate the arm as we expect extra
To maximize flexibility, both in the style and apparent motion of the DOFs to be added, reducing the effectiveness of the design. The
robot’s head, we exploited the relatively recent availability of low- arm is driven by anti-cogged brushless DC motors3 [9] through
cost portable projectors in a retro-projected animated face (RAF). a transmission consisting of a friction wheel pair and a timing
RAFs have been shown to be highly expressive, such as in Lighthead belt speed reduction. The whole arm module mounts the the spine
[3], Mask-Bot [10], Furhat [8], and Engineered Arts’ Socibot1 . via fasteners. The electronic diagram for this module is defined in
The RAF consists of a small projector, a lens or mirror, and a Figure 10.
projection surface. Quori’s head module utilizes a domed mirror Notable features of the arm design include access to and resolu-
and pico projector to warp a projected image onto a specially coated tion of the joint positions, drive motor abilities, and general safety
sphere (Figure 3). A rigid connection between the projector and pro- considerations. The approximate resolution of the joint position
jection surface has been designed to keep the projector and mirror
2 The projector is rated to provide 300 lumens, last 20,000 hours, and have native
aligned and minimize noise caused by the robot’s vibration. These
1280x720 HD resolution; however, there are only 132 lumens available to the spherical
components fit within a compact space approximately the size of surface since the only image reflected to the spherical head is a circle inscribed inside
the projection rectangle.
1 https://www.engineeredarts.co.uk/socibot/ 3 http://iq-control.com/
Preliminary system and hardware design for Quori,
a low-cost, modular, socially interactive robot HRI ’18 Workshop on “Social Robots in the Wild”, March 5, 2018, Chicago, IL, USA

Figure 4: Arm module CAD showing inside view of the trans-


mission; belts not shown. Figure 5: Holonomic base hardware schematic. Mounting is
for upper body or customized hardware. DoF axes driven by
the three actuators are highlighted, M1 and M2 are driven by
sensors are 0.022 degrees and 0.075 degrees for the shoulder joint the drive motors, and MT is driven by the turret motor.
and the drive motors, respectively. Access to both positions allow
us to check for slippage in the friction wheel or timing belt stages,
(ADA 303.2), 1/2-inch floor gaps (ADA 302, 407.4.3), and 1:12 inclines
as well as perform automatic calibration on boot-up of the system.
(ADA 405.2), which are common in real world deployments. The
The arm motors can produce approximately 0.3 Nm torque and are
max speeds the base is able to achieve are 0.8m/s in a straight line
able to rotate at approximately 16 revolutions per second, resulting
and 180 degrees per second with the turret. The field of view (FoV)
in possible shoulder joint speeds of about 1.2 radians per second.
of the laser distance scanner is shown in Figure 9 and was achieved
The arm design is expandable; we designed access for power and
by placing the sensor to maximize FOV, as shown in the plot. The
communication for further joints in the arms, such as an elbow,
on-board computer was selected to allow additional sensors to
while allowing the arm to rotate continuously.
be added that may require substantial computational resources.
We employed the following safety measures: setting a torque
Finally, the design allows the base to act as a stand-alone module
limit on the drive motors; a low-mass, inertial, and stiffness arm
independent of the upper-body humanoid torso.
mechanism that is safer according to the Head Injury Criterion [13];
The choice of design for the holonomic base ensures notable cost
and a friction wheel designed to slip in case the motor generates
reduction over other options. For example, with three motors, our
too much torque. For safety purposes, the arms are programmed to
base uses fewer actuators than other designs that require four or
stop and coast to prevent self-collision.
more motors [4]. Other holonomic designs may involve using an
Our primary goals in arm design are to ensure safe and precise
omniwheel or additional motors; however, they often suffer from
fluid motion at low cost. To achieve these goals, we used light-
performance drawbacks, such as vibration or complexity [5]. The
weight limbs with no payload capability and anti-cogging control
manufacturing of the base is made more affordable by utilizing
to make low-cost motors usable [9]. Arms that would be expected to
laser cut parts from sheet ABS and commercial off the shelf (COTS)
lift, push, or pull would need structural stability that typically leads
parts for the majority of the components, requiring only one major
to heavier and more expensive designs. Furthermore, a heavier arm
3D-printed part and two machined parts.
requires larger, and thus more expensive, motors to move. Low-cost
motors or servos could be used, though at the expense of precision,
as noted in [9] for the case of brushless DC motors.
3.4 1-DOF Spine
The spine is designed to not only support the arms and head (Figure
3.3 Holonomic Mobile Base 6, left), but also to produce useful and natural forward/backward
postural motion, (Figure 6, right). The spine’s 1-DOF allows the
Our mobile base module (Figure 5) is holonomic and is inspired
robot to express levels of engagement or disengagement by leaning
by the RAMSIS II design [5]. It is optimized for mobility using the
forward or backward. A secondary benefit of this design is its
design tools from Costa and Yim [1]. The base has three actuators
potential to minimize vibration generated during the motion of the
(Figure 5) that generate base linear and angular velocities in the
mobile base, leading to fluid, natural, and appealing motion that
ground plane and orient the upper body of the robot. Two casters
can be tuned in a way similar to that of a metronome. The arm and
serve to support the weight of the robot and increase the support
head modules connect to the spine and swing with it. The batteries
polygon. Power is delivered to the base via a slip ring between
and main on-board computer are stored within the spine structure.
the turret and differential drive base. Communication and control
The spine allows for easy attachment of additional custom hard-
of the base are achieved wirelessly via an Up Core single-board
ware, specifically, the arms and head. A new head module can easily
computer 4 . Space and USB ports are available for a laser scanner
be securely attached to the spine using the mounting holes pro-
or camera. The electronic diagram for this module is defined in
vided. The arms have similar mounting possibilities with the ability
Figure 10.
to screw into the spine column. There are also considerations for
The base measures 19 inches in diameter and 8 inches in height
shelves/ledges to be added to the spine for sensors, tablet, tray,
and complies with the 2010 Americans with Disabilities Act (ADA)
container mounting, etc.
Standards for Accessible Design; it can traverse over 1/4-inch bumps
Considerable space is allotted for the batteries and computer, 6.75
4 https://up-shop.org/up-boards/140-up-core-4gb-32gb-emmc-memory.html
x 6.0 x 8.4 inches and 8.0 x 8.0 x 2.9 inches, respectively. Currently,
HRI ’18 Workshop on “Social Robots in the Wild”, March 5, 2018, Chicago, IL, USA A. Specian et al.

curved surfaces and eased edges that facilitate acceptance. The


overall form is a body, or envelope, that is made of large parts with
consistent features: the spherical head is connected to the rounded
torso by a stalk. The outsized and softly curved forearms meet the
torso by a slender femoral shaft. The upper body is supported over
the base by a geometrically simple waist. This yields a perception
of a network of discrete, soft spheroids connected by simple masts,
rather than a body that is blob-like or mechanical across its surface.

4.2 Manufacturing Panels


Figure 6: Upper torso and spine hardware schematic (left).
We are currently working from the principles mentioned in Sec-
Extreme positions the robot achieves by bowing forward 30
tion 4.1 toward an attainable price point for fabrication that also
degrees (right) or leaning back 15 degrees (middle) which are
allows user choice in the robot embodiment details. Our panels are
mechanically limited to prevent collisions.
designed to be swappable for different colors or materials (Figure
7). The panels are easily removed or replaced via the magnetic
and mechanical alignment and securing features, avoiding visibly
we fit 40-ampere-hours of sealed lead acid batteries that power mechanical buttons or fasteners on the surface. Users also have the
the whole robot. While many options exist for small form factor option to perform more involved exterior design changes similar to
computers, we have ensured sufficient space for a computer with ONO [11], such as completely changing the style of the robot. We
considerable computational resources for in the wild operations, used 3D printing, manufacturing customizable panels from CAD
such as an Alienware Alpha R2 or NVIDIA Jetson Tk1. printable panels to fit the spine.
The spine design is low cost because of its simple structure man-
ufactured from laser cut sheet material and a handful of additional
hardware. Cost, as well as weight, are saved by using the batteries
as counter-mass for the spine.

4 DESIGN AND MANUFACTURING


4.1 Appearance Design
Physical appearance is a key attribute for a robot designed for social
interaction. We continue to discover what modular appearance
accessories and baseline physical characteristics address the needs
of the HRI community (as indicated by community input) while
also retaining stylistic consistency when appropriate. A sample
design for Quori is shown in Figure 1.
Quori has undergone an “envelope” design development. The
underlying robot and mechanical systems are clad with a panelized Figure 7: Basic design considerations and features for attach-
torso, base, and arms. This method allows for the working apparatus ing Quori panels. Removable panels not shown to demon-
of the robot and the body to be separable. The design for each body strate ease of access to chest, battery, and computer areas.
part was tested to allow for local freedom of movement. For example,
the torso was designed with collision detection for the movements
of the arm at the shoulder as well as for the rotations at the waist.
Each panel was considered not only in appearance, fit, and finish, 4.3 Customization for Research Goals
but also in terms of easy disassembly for access to the robot. The The following is a list of possible Quori customizations.
base panel curves are designed to increase the distance between
the user and the robot for both safety and social proxemics [6]. • Replace head module with a tablet or place one in the chest
Significant attention was given to the holistic appearance of • Hold objects in the arms or on a tray attached to the spine
Quori. The torso, arms, and base were designed to be an identifiable, • Upgrade computation capabilities
self-consistent whole: color, seaming, and surface curvature are con- • Remove or replace the arms
tinuous among the parts. These features (seaming, curvatures, etc.) • Detach the mobile base, head, torso, or arm modules.
of the body were designed to address specific community-identified
HRI issues of gender, the Uncanny Valley [7], and acceptance. The 5 CONCLUSIONS
gender identity is meant to be dampened without being generic, This paper introduced the design of Quori, a low-cost socially in-
the size and appearance is slightly abstract to not mimic human teractive robot platform. We described the features and utility of
physiognomy and therefore avoid the Uncanny Valley, and the the four modules and identified decisions for ensuring a low-cost
geometry of the robot is meant to be recognizably friendly — we design. The modules are designed to assemble together into an
avoided sharp corners or threatening musculature in favor of softly aesthetic design that meets issues identified by the HRI community.
Preliminary system and hardware design for Quori,
a low-cost, modular, socially interactive robot HRI ’18 Workshop on “Social Robots in the Wild”, March 5, 2018, Chicago, IL, USA

ACKNOWLEDGMENTS
This work is supported by the National Science Foundation under
Grant No. CNS-1513275 and CNS-1513108.

REFERENCES
[1] John Tighe Costa and Mark Yim. 2017. Designing for uniform mobility using holo-
nomicity. In Robotics and Automation (ICRA), 2017 IEEE International Conference
on. IEEE, 2448–2453.
[2] Frédéric Delaunay and Tony Belpaeme. 2012. Refined human-robot interac-
tion through retro-projected robotic heads. In 2012 IEEE Workshop on Advanced
Robotics and its Social Impacts (ARSO).
[3] Frédéric Delaunay, Joachim De Greeff, and Tony Belpaeme. 2009. Towards
retro-projected robot faces: an alternative to mechatronic and android faces. In
Robot and Human Interactive Communication, 2009. RO-MAN 2009. The 18th IEEE
International Symposium on. IEEE, 306–311.
[4] Travis Deyle. 2010. VolksBot Omni: A Holonomic Robot Base Using only Robo-
tis Dynamixel Servos. (Aug 2010). http://www.hizook.com/blog/2010/08/01/
volksbot-omni-holonomic-robot-base-using-only-robotis-dynamixel-servos
[5] Ahmed El-Shenawy, Andrea Wellenreuther, Andre S Baumgart, and Essam
Badreddin. 2007. Comparing different holonomic mobile robots. In Systems,
Man and Cybernetics, 2007. ISIC. IEEE International Conference on. IEEE, 1584–
1589.
[6] Ross Alan Mead. 2015. Situated Proxemics and Multimodal Communication: Space,
Speech, and Gesture in Human-Robot Interaction. Ph.D. Dissertation. University
of Southern California.
[7] Masahiro Mori. 1970. The uncanny valley. Energy 7, 4 (1970), 33–35.
[8] Samer Al Moubayed, Gabriel Skantze, and Jonas Beskow. 2013. The FurHat
Back-Projected Humanoid Head–Lip Reading, Gaze and Multi-Party Interaction.
International Journal of Humanoid Robotics 10, 01 (2013), 1350005.
[9] Matthew Piccoli and Mark Yim. 2016. Anticogging: Torque ripple suppression,
modeling, and parameter selection. The International Journal of Robotics Research
35, 1-3 (2016), 148–160.
[10] Brennand Pierce, Takaaki Kuratate, Christian Vogl, and Gordon Cheng. 2012.
âĂIJMask-Bot 2iâĂİ: An active customisable robotic head with interchangeable
face. In Humanoid Robots (Humanoids), 2012 12th IEEE-RAS International Confer-
ence on. IEEE, 520–525.
[11] Cesar Vandevelde and Jelle Saldien. 2016. Demonstration of OPSORO-an open
platform for social robots. In Human-Robot Interaction (HRI), 2016 11th ACM/IEEE
International Conference on. IEEE, 555–556.
[12] John P Whitney and Jessica K Hodgins. 2014. A passively safe and gravity-
counterbalanced anthropomorphic robot arm. In Robotics and Automation (ICRA),
2014 IEEE International Conference on. IEEE, 6168–6173.
[13] Michael Zinn, Oussama Khatib, Bernard Roth, and J Kenneth Salisbury. 2004.
Playing it safe [human-friendly robots]. Robotics & Automation Magazine, IEEE
11, 2 (2004), 12–21.
Figure 9: Top: plot showing the location required to have a
A ADDITIONAL FIGURES maximum FOV as measured by: the largest continuous visi-
bility (yellow), total visibility near the base (orange), and to-
tal visibility near the limit of the sensor (blue). Bottom: FOV
of our current design with the sensor offset 100mm. Blind
spots shaded in gray and the outer circle marking an 8-meter
radius about the robot marked as the smaller inner circle.

Figure 8: A sample set of eyes.


HRI ’18 Workshop on “Social Robots in the Wild”, March 5, 2018, Chicago, IL, USA A. Specian et al.

Figure 10: Power and electronic component diagram for Quori showing power (purple), actuation (red), computation (green),
sensing (yellow).

You might also like